Implementing a GRU Neural Network for Flood Prediction in Ashland City, Tennessee

TL;DR

This study develops and evaluates a GRU neural network model for flood prediction in Ashland City, Tennessee, using water level data, achieving high accuracy and demonstrating its potential for disaster management.

Contribution

The paper introduces a GRU-based flood prediction model tailored for Ashland City, utilizing real-time water level data and demonstrating high predictive accuracy.

Findings

The model explained 98.2% of the variance in water levels.

The GRU model showed high accuracy with low error metrics.

Potential for improved flood preparedness in Ashland City.

Abstract

Ashland City, Tennessee, located within the Lower Cumberland Sycamore watershed, is highly susceptible to flooding due to increased upstream water levels. This study aimed to develop a robust flood prediction model for the city, utilizing water level data at 30-minute intervals from ten USGS gauge stations within the watershed. A Gated Recurrent Unit (GRU) network, known for its ability to effectively process sequential time-series data, was used. The model was trained, validated, and tested using a year-long dataset (January 2021-January 2022), and its performance was evaluated using statistical metrics including Nash-Sutcliffe Efficiency (NSE), Root Mean Squared Error (RMSE), Percent Bias (PBIAS), Mean Absolute Error (MAE), and Coefficient of Determination (R^2). The results demonstrated a high level of accuracy, with the model explaining 98.2% of the variance in the data. Despite minor discrepancies between predicted and observed values, the GRU model proved to be an effective tool for flood prediction in Ashland City, with potential applications for enhancing disaster preparedness and response efforts in Ashland City.